A Guth Labs publication ↗
Back to News

Cloudflare brings multimodal EmbeddingGemma 2 to Workers AI

Cloudflare brings multimodal EmbeddingGemma 2 to Workers AI

November 12, 2026 | guthlabs.newsroom

Cloudflare has added Google's EmbeddingGemma 2 to Workers AI, giving developers a hosted embedding option for mixed collections of text, code, images, video and audio. The announcement is a new Cloudflare availability event, not evidence that Google first released the underlying model today.

The model represents those inputs in a shared 768-dimensional vector space. Cloudflare also documents shorter 128-, 256- and 512-dimensional outputs, trading some retrieval quality for smaller vectors. Its context window is 8,192 tokens across the content in an input. These are provider-described capabilities, not results of a Guth benchmark.

The Workers AI model page identifies the endpoint as @cf/google/embeddinggemma-2 and lists a price of $0.0017 per million input tokens. It documents text requests and mixed-media messages, with media supplied as base64 data URIs. This is an embedding price, not a quote for an entire search application.

Why it matters for the fleet: analysis

A shared embedding space could help a product search documents, screenshots and recorded material through one retrieval system instead of treating each medium as a disconnected library. For Guth, the useful experiment would be whether a text question retrieves the right source across those formats, with citations preserved.

The migration should not be treated as changing an API string. Test a separate index, compare retrieval on representative questions, and measure indexing, storage and query costs together. A smaller vector is useful only if the relevant source still appears reliably. Keep the existing index available until the alternative passes those checks.

This also complements, but does not establish automatic integration with, the separately announced D1 source for AI Search. Workers AI model availability alone does not prove that every Cloudflare retrieval product accepts the model as a drop-in choice.

Verification and limits

The November 12 changelog and complete model documentation were read directly. No exact publication time was displayed, so this is not labeled a last-hour launch. No live inference, independent benchmark or Guth deployment was performed. Earlier accessible coverage mentions EmbeddingGemma 2 in research context; no Workers AI availability story was found. Unrecovered prior drafts prevent an absolute completeness claim.

Cited primary evidence

Related Guth coverage

  • cloudflare-ai-search-d1-beta-2026-11-12: separate D1 retrieval-source announcement, covered separately.

Independent corroboration

None established for this deployment's performance. Provider documentation is the evidence for announced availability and specifications.

Sources and citations

Revision history

Published .